The Research Of J Wave Classification Method Based On Multi-feature Fusion

J wave is a congested waveform,which generates near the junction of the end of QRS complex and the beginning of ST segment in electrocardiogram(ECG).It is approved in clinic,the variation of J wave can cause arrhythmia,sudden cardiac death or other cardiovascular diseases,which seriously threatens human life.Therefore,it is of great significance to improve the diagnostic efficiency of J wave related diseases and reduce the mortality of these pathema by means of technology.To solve the above problem,this paper applies machine learning technology to achieve automatic classification of J wave signals.The main contents of this paper are as follows:(1)Based on multi-scale recursive analysis in time-frequency domain,put forward the J wave detection method to realize the high-accuracy identification of J beats in ECG.In this study,after obtaining raw data from cooperative hospital and completing the operations of database construction,preprocessing and so on,firstly,tunable-Q wavelet transform(TQWT)is used to realize the multi-scale time-frequency decomposition of the signals.Then,the dynamic recurrence quantification analysis method is used to extract the quantitative features that represent the similarity of the decomposition coefficients.Subsequently,the t-test and sequential floating forward selection(SFFS)algorithm is combined to complete feature selection.Ultimately,the least squares support vector machine(LS-SVM)classifier which optimized by the bacterial foraging optimization algorithm(BFOA)is served to perform efficient recognition of normal beats and J wave beats.(2)Based on multi-layer classifier and multi-angle features fusion,a classification method is proposed to distinguish the normal beats,benign J wave beats and malignant J wave beats.For the purpose of getting more abundant and comprehensive signal expression,this method firstly obtains different features from the time domain,frequency domain and time-frequency domain.Later,the feature optimization and weighted fusion technique is used to combine the above multi-domain features into the final feature sets.And in the subsequent classification stage,taking the previous final feature set as the input vectors,the two layer classification framework which composed of LS-SVM is constructed,where the first layer classifier is used to realize the pre-detection of the J wave,and the second layer classifier is employed to distinguish the benign J wave and the malignant J wave.The experimental results show that the average accuracy of the J wave detection is 97.2%.At the same time,more than 80% classification accuracy is also obtained in the distinction of benign J wave and malignant J wave.Therefore,this algorithm can be used as an effective J wave aided discriminationmethod,which can provide some help for the diagnosis of J wave related diseases in clinical medicine.